How to Choose an AI Consultant: Red and Green Flags
November 18, 2026

Every week I talk to a business owner who just paid an AI consultant and got very little for it. Sometimes the story is a $20,000 strategy deck that nobody ever used. Sometimes it is a chatbot that answers three questions and breaks on the fourth. Sometimes it is a six-month project that is still, somehow, in the discovery phase.
Here is the uncomfortable part: in most of those stories, the consultant was not a scammer. They were just the wrong consultant for that business, hired the wrong way, for the wrong job. And the owner did not have a way to tell the difference before signing.
I have sat on both sides of this table. I build AI agents for businesses, and I also watch owners get burned by bad AI advice. So let me give you the honest version of how to choose, the way I would explain it over coffee: the red flags that should stop you cold, the green flags worth paying for, the questions to ask on the first call, and what a good engagement actually looks like once the work starts.
Start with the job, not the vendor
Before you talk to anyone, write down one sentence: the specific job you want AI to do. Not "use AI in my business" and not "become an AI-powered company." Something like: "Answer the calls we miss after 6pm and book the appointment." Or: "Stop my team retyping job details from email into our scheduling system." Or: "Answer the 20 questions every customer asks before they buy."
That one sentence does more work than any vendor comparison. A consultant who is right for you will grab that sentence and start asking questions about it: how many calls, what kind of jobs, what happens when the booking is wrong. A consultant who is wrong for you will nod politely and steer the conversation back to their platform, their framework, or their slide deck.
The sentence also protects you from the most common trap in AI consulting, which is buying a solution and then going looking for a problem it solves. If you do not know the job first, every consultant sounds convincing, because every consultant has a tool that does something impressive in a demo. The demo is not the product. The product is what happens on a Tuesday afternoon when a real customer calls with a messy, badly worded request.
So write the sentence down. Then, and only then, start taking calls. If a consultant cannot connect what they sell to that sentence in the first ten minutes, you have your answer, and it did not cost you anything.
The red flags
Let me start with the warnings, because they are cheaper to learn from reading than from experience.
They promise a specific return before seeing your business. "We will save you 40 percent on staffing costs" or "This pays for itself in six weeks" sounds great in a sales call, and it is meaningless until someone has looked at your call volume, your margins, and your team. A consultant who guarantees numbers before doing any discovery is either guessing or lying, and neither one is a good foundation for a five-figure engagement. Real ROI talk starts with your numbers, not theirs.
They talk about tools, never about your workflow. If the first meeting is a tour of features, model names, and integrations, and nobody asks how your team actually handles a customer today, walk away. The tool is the easy part. The hard part is the workflow: who does what now, where it breaks, what a mistake costs you. A consultant who skips that conversation will build you something technically impressive that your team quietly stops using within a month.
They cannot show you anything working. Not a slide, not an architecture diagram, not a screenshot. A working agent, on a real business, handling a real task. Ask to see it, and ask to try breaking it. Every honest consultant has something live they can show you, even if it is a small pilot for a past client. If everything is "in development" or "under NDA," that usually means there is nothing to show.
They want a long contract before any pilot. A 12-month commitment, paid upfront, before you have seen a single thing run in your business, is a red flag the size of a billboard. Good consultants are confident enough to start small, because a small pilot that works sells the bigger engagement better than any contract ever could. The standard shape is a paid pilot of about 30 days, then a decision, then a larger scope. Anyone who refuses that shape is telling you they are not confident the pilot would work.
They cannot name a failure. Ask what went wrong on their last project. Every real project has something: a workflow that was harder than expected, a client whose team resisted the change, a pilot that had to be rebuilt. A consultant who says everything always goes smoothly is either inexperienced or dishonest. The good ones will tell you the story unprompted, because they learned something from it, and that learning is part of what you are paying for.
They disappear after the sale. Some consultants sell the project, then hand it to a junior team you never met, or to an offshore contractor, and you only hear from them again at the invoice. Ask in the first call who will actually do the work, how often you will talk to them, and what happens when something breaks on a Friday evening. If the person selling you the project will not be the person building it, you are not hiring an expert, you are hiring a sales team.
The green flags
Now the good signs, the ones worth paying for.
They ask about your numbers first. The best first question a consultant can ask is not "what AI tools have you tried" but "how many calls do you miss in a week, and what is a job worth to you." That question means they are thinking about your business, not their product. It also means they will be able to tell you honestly whether the project is worth doing at all, and a consultant who talks you out of a bad project is worth more than one who sells you a good one.
They show you something working, then let you poke at it. A live demo with a real use case, where you can ask the awkward questions and watch it handle them, or fail gracefully when it cannot. This is the single most useful thing you can do in a first meeting. I would trust a rough working demo over a polished slide deck every time, because a demo cannot hide the things a deck can.
They talk about handover and training from day one. The project is not done when the agent works. It is done when your team can run it, fix the small things, and spot the big things before they become problems. A consultant who plans the training and the handover in the first conversation is thinking about your business after they leave, which is exactly the consultant you want. Ask what happens on day 31, when the pilot ends and they are not in the room anymore.
They price the pilot separately from the build. A small, fixed-price pilot with a clear deliverable and a clear decision point is the greenest flag there is. It says: I am confident enough to prove this works before asking for real money. The pilot also gives you something invaluable, which is a low-cost way to learn how this consultant actually works, how they communicate, and whether your team likes working with them.
They name the risks without being asked. Data quality, team adoption, edge cases the agent will get wrong, the cost of keeping it running. A consultant who brings up the risks is not trying to talk you out of the project, they are showing you they have done this before and know where it breaks. The one who says there are no risks is the one who has not done it before.
Their references are real people you can call. Not logos on a website, not written quotes, but two clients who will pick up the phone and talk to you. And when you call them, they sound like real businesses with real opinions, including complaints. I will say more about how to run those calls below, because it is the highest-value hour you will spend in this whole process.
Red flags vs green flags, side by side
Here is the whole guide in one table. If you remember nothing else, remember this: the red flags are all about the consultant protecting themselves, and the green flags are all about the consultant protecting you.
| What to watch | Red flag | Green flag |
|---|---|---|
| First meeting | A tour of their tools and features | Questions about your numbers and workflow |
| Promises | A guaranteed ROI before any discovery | A range, based on your actual volumes |
| Proof | Slides and screenshots only | A live demo you can try to break |
| Contract | 12 months paid upfront, no pilot | A paid 30-day pilot, then a decision |
| Failures | Everything always goes smoothly | An honest story about what went wrong |
| The team | You never meet the people doing the work | You talk to the builder from day one |
| Risks | No risks mentioned, ever | Risks named before you ask |
| Handover | Training is an add-on, or never mentioned | Training and handover planned from the start |
| References | Logos and written quotes only | Two clients who will take your call |
Read the table and notice the pattern. None of the green flags are about technology. They are all about honesty, proof, and who carries the risk. That is the real test. The technology changes every few months anyway. The character of the person you hire does not.
The questions to ask on the first call
You do not need a long checklist. Five questions, asked plainly, will tell you almost everything.
What would you build first, and why that instead of the other options? You are testing whether they can prioritize. A good consultant will have a clear answer and a reason tied to your business: the missed calls are worth more than the website chat, or the retyping problem is cheaper to fix than the reporting problem. A weak consultant will say everything is important, which is another way of saying nothing is.
What does the pilot look like, and what do I get at the end of it? You want a specific deliverable: one workflow, running, with your team using it. If the answer is a strategy document or a roadmap, the pilot is not a pilot, it is a paid sales process.
Who does the work, and how often will I hear from them? You want names, not roles. You want to know if the person on this call is the person building the thing, and what a normal week looks like: a short call, a written update, a shared channel where your team can ask questions.
What happens when it breaks? Because it will break, at some point, on some edge case nobody predicted. The answer you want is a real one: who notices, how fast, and what it costs you. "It will not break" is the wrong answer, and it is also a lie.
What would make you tell me not to do this project? This is my favorite question, because it forces honesty. A good consultant has an answer ready: your volume is too low, your data is too messy, your team is not ready. A bad consultant will struggle, because they have never talked a client out of a sale.
Ask those five, then listen. Not just to the answers, but to how comfortable the consultant is with being questioned. The best ones enjoy it. They have answered these questions before, and they know that a client who asks hard questions before signing is a client who will be easy to work with after signing.
What a good first 30 days looks like
Once you have picked someone, here is what the first month should feel like. If it does not look like this, say so early, because the first 30 days set the pattern for everything after.
About a quarter of the time goes to discovery, and I mean real discovery: sitting with your team, watching how they actually work, reading the actual emails and call notes, not just interviewing the owner. Then the biggest chunk, roughly 40 percent, goes to building the pilot on one real workflow. Training your team takes about a fifth of the time, and it matters as much as the build, because an agent nobody on your team understands is an agent nobody trusts. The rest is measuring what happened and handing the whole thing over cleanly.
How a good consultant spends the first 30 days: most of the time building one real thing, with real discovery before it and real training and measurement after it. If a proposal spends half the month on strategy documents, it is not a pilot.
The key word in all of this is "one." One workflow, one pilot, one decision at the end of the month. Consultants who try to fix five things at once usually fix none of them properly, because their attention is split and your team cannot absorb five changes at once. The most common first pilot I see is a voice agent that answers missed calls, because the problem is easy to measure and the payoff is obvious, but the shape works for any workflow: research that eats your team's mornings, follow-ups that never get sent, quotes that take two days to go out. Pick the one that hurts most, prove it works, then decide what comes next.
How much should it cost
Let me be direct about money, because vague pricing is how owners get hurt.
A paid pilot should run about $1,500 to $2,500 for 30 days. That is enough for a consultant to do real discovery and build something working, and small enough that you are not betting your quarter on it. If someone offers you a free pilot, be careful: free usually means a generic demo with your logo on it, not a pilot built on your business, and you will learn nothing about how they work.
A real implementation, where the consultant configures an agent for one part of your business, connects it to your systems, and trains your team, typically runs $5,000 to $15,000 in setup, depending on how messy your systems are and how many workflows are involved. The low end is a single clean workflow. The high end is several workflows with real integrations and real edge cases.
If they stay on to run it, improve it, and add the next workflow, expect a monthly retainer of $3,000 to $7,500. That sounds like a lot until you compare it to the cost of the problem: a missed-call problem alone can cost a service business thousands a month in lost jobs, and the retainer is the price of someone who keeps the system working and keeps making it better.
The numbers that should worry you are the ones far outside these ranges with no explanation. Far below usually means a resold tool, not consulting. Far above usually means you are paying for an agency's overhead, not for better work. In both cases, the right response is the same: ask them to explain the number, in plain English, tied to your business. If they cannot, the number is not real.
How to check references without getting played
References are the most valuable hour in this whole process, and most owners waste it by asking the wrong questions.
First, insist on two clients you can call. Not emails, not written quotes, a phone call. Any consultant with happy clients will arrange it, because their clients like them and are happy to say so. If arranging a call is somehow impossible, treat that as a red flag on its own.
When you get them on the phone, ask three questions. What exactly did the consultant build, in plain terms? What went wrong during the project? And would you hire them again for the next one?
The second question is the one that matters. Every real project has something go wrong: a timeline that slipped, a workflow that turned out to be harder than anyone thought, a team member who resisted the change. A reference who cannot name a single problem is either not a real client or was coached, and both are reasons to walk away. A reference who tells you what went wrong, and then tells you how the consultant fixed it, is the best reference you will ever get.
Also listen for what the reference does not say. If they describe the consultant in vague terms, "great to work with, very professional," but cannot describe what was actually built, the project was probably small or forgettable. If they can tell you the numbers, how many hours it saved, how many calls it now answers, how much faster quotes go out, that is a consultant who delivered something measurable.
One more thing: call the reference yourself, do not let the consultant set up a group call where they are on the line. People are honest on a one-to-one call in a way they are not when the consultant is listening.
The bottom line
Choosing an AI consultant is not really a technology decision. It is a people decision, and the test is simple: does this person ask about your business before pitching their product, can they show you something working, will they start small and prove it, and will they tell you the truth when the truth is uncomfortable.
The red flags are all versions of the same thing, a consultant protecting themselves: guaranteed numbers, long contracts, no live proof, no failures, no risks. The green flags are also all versions of the same thing, a consultant protecting you: honest questions, a working demo, a small pilot, named risks, real references.
If you find the green flags, the rest is straightforward. Start with one job, one pilot, one month. Measure it honestly. Then decide. That is how good AI projects start, and it is how you avoid becoming the owner in the story I opened with, the one who paid five figures and got a slide deck.
And if you want a second pair of eyes before you sign anything, that is exactly what AI consulting is for. I would rather spend an hour with you before you hire someone than help you untangle a bad engagement after the money is gone, and if we are not the right fit for the job, I will tell you that too.
Frequently asked questions
Want AI working for your business?
We help small and mid-size businesses put AI agents to work: voice agents, sales automation, and custom builds, plus training so your team actually uses it.
Talk to us about AI